Multi-step Ahead Power Demand Forecasting in Smart Grid

A precise prediction of the power grid variable is an essential step in electrical decision-making problems. The major aim of this study is to improve the capability of artificial intelligence (AI) specifically neural networks in forecasting the power grid variables. For this purpose, it is extended a new hybrid neural network in which pre-specified numbers of simulated data series are generated by the tuned ARMA and GARCH-type models along with other explanatory features including exogenous and endogenous variables which are considered as input variables. Moreover, the proposed neural network has been tuned by an efficient particle swarm optimization (PSO) algorithm to discover the best features and to more accurate forecasts of the power grid variable as the output. Using a real dataset, it is demonstrated how the proposed model and adding new features could reasonably improve the results of a traditional neural network concerning the different performance measures. Moreover, it is shown by the obtained results how the mentioned simulated data series as the new input to the neural network could enhance the accuracy of power grid forecasting.